Debug missing choices array treated as empty success in the CLI LLM summarizer

Do not start with a speculative fix for the CLI LLM summarizer. Force the failure on purpose, save the before output, then change one cause at a time. Retries are allowed only for transient conditions—not for bad input that will fail forever on llm-api-hello.

async function retry(call){for(let n=0;n<3;n++){try{return await call()}catch(e){if(e.status!==429||n===2)throw e;}}}
let n=0; retry(async()=>{if(n++<1)throw Object.assign(new Error('rate'),{status:429});return 'ok'}).then(console.log);

Expected evidence: ok after one simulated 429. If you cannot reproduce on demand, you do not yet control the failure mode for llm-api-hello.

After repair, rerun the exact reproduction command. Keep the failing fixture as a regression seed for the observability page. For the command-line LLM summarizer, remember the claim you are restoring: fake transport returns a summary string; real adapter is isolated behind env credentials.

Store a one-command reproduction for: missing choices array treated as empty success, or retries on 400. The command should use system+user job object for an embeddings definition sentence or a minimal mutant of it. Paste the failing output into notes/failure-before.txt (or your shell scrollback as copied text). After the fix, paste notes/failure-after.txt and keep both.

Retries belong only on transient faults. If the failure is bad input, a bad allowlist, or a logic bug in the CLI LLM summarizer, retrying will amplify cost without repairing trust around turn one user passage into a bounded one-sentence summary via an HTTP-shaped client.

Classify the failure as prevent, detect, contain, or recover—using this lab’s language, not a generic poster. For llm-api-hello, the first fix should usually be detect+prevent at the boundary, because missing choices array treated as empty success, or retries on 400 is cheaper to stop early than to explain in production prose.

At the debugging stage for llm-api-hello, the job is narrower than finishing a product demo. You are creating one progressive evidence piece about system+user job object for an embeddings definition sentence that later pages inherit without redefining success. Keep that fixture small enough to inspect by hand, keep outputs copy-pasteable as text, and refuse to narrate this baseline as if it were a production SLA: offline fake transport producing a fixed summary.

For this page specifically, success looks like before/after evidence for the characteristic failure while still centering the user decision to turn one user passage into a bounded one-sentence summary via an HTTP-shaped client. If you cannot point to a file, command, or assertion that proves that for the CLI LLM summarizer, stay on this page instead of advancing.

API: chat completions · Glossary: API key

Previous · Next